Journal of Experimental Psychology: Human Perception and Performance 1995, Vol. 21, No. 3, 584-601
Copyright 1995 by the American Psychological Association, Inc. 0096-1523/95/$3.00
Automatic Access to Object Identity: Attention to Global Information, Not to Particular Physical Dimensions, Is Important Muriel Boucart
Glyn W. Humphreys University of Birmingham
Centre National de la Recherche Scientifique Universite Rene Descartes Paris V
Jean Lorenceau Centre National de la Recherche Scientifique Universite Rene Descartes Paris V The authors examined whether, by attending to physical properties of objects, participants can prevent the activation of semantic information. Participants received a reference object followed by a display containing both a matching target and a distractor. In Experiments 1 and 2, participants attended to motion and to surface texture, respectively. Some evidence for the processing of semantic information occurred. This result contrasted with a previous study in which no evidence for semantic information processing was apparent in a color matching task (M. Boucart & G. W. Humphreys, 1994). In Experiment 3, pictures were used with outline contours composed of randomly distributed red and green dots, one color being overrepresented. Participants matched pictures according to the dominant color. Evidence for semantic processing emerged. The authors suggest that these results cannot be explained in terms of attention operating differently on separate physiological channels. Instead it is proposed that what is crucial in activating stored object representations is whether the global configuration of the picture is processed.
The question of whether study participants can attend to physical properties of stimuli and in so doing filter out high-level information has a long history in the field of auditory perception. Despite this long history, a definitive answer remains elusive. Some studies report effective lowlevel filtering (e.g., Broadbent, 1958), others report minimal effects of attention to a physical property on preventing high-level processing (e.g., Moray, 1959). Some of the more recent studies (Johnston & Heinz, 1978) suggest that a definitive answer may not even be possible. Whether physical filtering is effective depends on the difficulty of the filtering task. Stronger evidence for high-level processing of filtered stimuli emerges when filtering is difficult. In vision research, relatively few studies have addressed the same question. For instance, many studies of selective attention in vision have required participants to attend to a physical property of the stimulus (e.g., its color in the Stroop paradigm, McLeod, 1991; its spatial location in the
Eriksen interference paradigm, Eriksen & Eriksen, 1974). Participants are then required to retrieve identity or name information from the attended target. Under such circumstances semantic or identity information from unattended elements affects performance. However, such studies do not assess whether access to high-level information may be prevented when the task only requires the processing of a physical property of the target. This last situation was investigated by Boucart and Humphreys (1992, 1994). They used a matching task based on a physical property of the form (e.g., its shape, orientation, or size) and manipulated the semantic relations between the stimuli. A reference object was displayed centrally and followed by two laterally presented objects: a target and a distractor. In different experiments, participants were asked to decide which of the two lateral objects (left or right) had (a) the same global shape (round or oval) as the reference picture (Boucart & Humphreys, 1992), (b) the same global orientation (oblique or horizontal) as the reference picture, or (c) the same size (large or small) as the reference picture (Boucart & Humphreys, 1994), regardless of the identity of the objects. Semantic effects on performance were assessed by manipulating the semantic relations between the reference object, the target, and the distractor. The reference object and the target (the lateral figure having the same shape, orientation, or size as the reference) were either physically identical, semantically related, or semantically unrelated. For each of the three target conditions, the reference object and the distractor (the lateral picture having a shape, orientation, or size different from that of the reference) were either semantically related or unrelated. Seman-
Muriel Boucart and Jean Lorenceau, Laboratoire de Psychologic Experimentale, Universite Rene Descartes Paris V, Centre National de la Recherche Scientifique, Paris, France; Glyn W. Humphreys, Centre for Cognitive Sciences, School of Psychology, University of Birmingham, Birmingham, England. This work was supported by a grant from the Medical Research Council of the United Kingdom. Correspondence concerning this article should be addressed to Muriel Boucart, who is now at Hopitaux Universitaires de Strasbourg, Institut National de la Sante et de la Recherche Medicale (U405), De>artement de Psychiatrie I, 67091 Strasbourg, France. Electronic mail may be sent via Internet to
[email protected].
584
ATTENTION TO GLOBAL INFORMATION tic processing was evidenced by (a) shorter response times (RTs) and a lower error rate when the target was semantically related to the reference object, relative to when these items were semantically unrelated; and (b) longer RTs and a higher error rate when the distractor was semantically related to the reference object, relative to when the two pictures belonged to different semantic categories. In contrast, there were no effects of the semantic relations between the stimuli when meaningless forms closely matched to the meaningful pictures were used as stimuli (Boucart & Humphreys, 1992). These results suggest that semantic information affected performance, even though the tasks required attention to a physical property of the stimuli. Logically, physical properties of visual stimuli must be processed before semantic information is contacted (though physical processing needs not be completed before access to semantics takes place; see Marcel, 1983). Also, physiological studies show that physical dimensions of visual stimuli are extracted and integrated at different levels of representation (Desimone & Ungerleider, 1989; Livingstone & Hubel, 1987; Maunsell & Newsome, 1987; Zeki, 1990a) before access to semantic or name information takes place (Bonnet, 1989; Riddoch & Humphreys, 1987). Several studies have investigated whether different physical properties such as size, orientation, and color affect object recognition. For instance, Paivio (1975) manipulated size in a matching task in which participants were asked to decide which of two simultaneously presented objects was larger in real life. Faster RTs were found when depicted and real sizes were congruent, relative to when they were incongruent. In a recognition memory task, Jolicoeur (1987) observed faster RTs when a target object matched the size of the same object shown in a learning phase relative to when the target and the learned objects differed in size. However, Biederman and Cooper (1992) found no effect of size on the magnitude of priming in a repetition priming task, and Ellis, Allport, Humphreys, and Collis (1989) also observed size constancy in a matching task requiring name matching of two objects varying in size. The depicted orientation of objects has also been found to affect both naming times and the discrimination of whether features were at the top or bottom of objects (Jolicoeur, 1985; McMullen & Jolicoeur, 1992; Maki, 1986). Furthermore, variations in viewing angle affect both matching (Ellis et al., 1989) and object identification performance (Palmer, Rosch, & Chase, 1981). Data are less clear concerning whether physical properties involving surface information (color, luminance, texture) about objects affect their processing. Ostergaard and Daviddoff (1985) found no differences in performance between correctly and incorrectly colored drawings of objects in either a naming or a semantic categorization task (with small numbers of objects). Biederman and Ju (1988) observed an advantage for outline drawings over colored photographs of objects at a short exposure time (50 ms) and no difference between the two types of stimuli with a longer presentation time. In contrast, Price and Humphreys (1989) found an advantage when participants had to correctly name or categorize colored photographs of objects, relative to
585
when incorrectly colored line drawings, black and white line drawings, or black and white photographs were used. Such results demonstrate that object recognition can be directly affected by manipulations of the visual attributes of objects. Indeed, luminance (e.g., shading) or texture can guide object recognition, for example, by helping the computation of boundaries of objects in visual scenes (Cavanagh & Leclerc, 1989; Vaina, 1987). In general, such data showing effects of physical properties on object recognition are consistent with these physical properties of objects being processed before access to semantic or lexical representations takes place. In contrast to this, Boucart and Humphreys's (1992) results indicate that although physical information may be processed first, it may not always be attended selectively without semantic processing taking place. It is interesting though that Boucart and Humphreys (1994) found that this did not hold for all physical properties of objects. It held when participants attended to the global shape of the object (Boucart & Humphreys, 1992) and when they attended to its size or global orientation (Boucart & Humphreys, 1994). However, there was no evidence of semantic processing when the matching task was based on the color (red and green) or the luminance (black and dark grey) of the pictures. RTs and the error rate were the same whether or not the reference object and the matching target were physically identical or semantically related. This was not simply a question of task difficulty. Semantic effects on task performance were apparent in an orientation matching task matched for difficulty to a task requiring matching on the basis of the luminance of the stimuli. Boucart and Humphreys (1994) proposed three accounts of why luminance and color could be attended selectively, whereas form properties (such as global shape, orientation, or size) were not. A first explanation was in terms of separate processing channels for form and for color or luminance information (Livingstone & Hubel, 1987, for a review). Color and luminance may be attended selectively because their processing is mediated through separate neural channels to form. However, this physiological explanation cannot account fully for the results, because it cannot explain why semantic interference emerged on tasks requiring attention to form information (global shape, orientation, and size matching). There still needs to be an account of attentional failure in these instances. Moreover, at higher levels of the visual processing there may be some integration of form and color information (Tanaka, Saito, Fukada, & Moriya, 1991). The second explanation related to the differential role of form and surface information in object recognition. As mentioned previously, several investigations suggest that surface information, such as color, luminance, or texture, is less efficient than form information in activating semantic representations of objects (Biederman & Ju, 1988; Ostergaard & Davidoff, 1985; Price & Humphreys, 1989). Consequently, surface properties may be attended without semantic information being accessed. Unfortunately, Boucart and Humphreys's experiments did not test the role of surface information directly because the stimuli were defined only by the luminance or the color of their outline contour.
586
M. BOUCART, G. HUMPHREYS, AND J. LORENCEAU
The third explanation concerned the different mode of processing of the pictures in the form matching tasks (based on global shape, orientation, or size) and in the color or luminance matching tasks. Semantic interference may have occurred in the form-based tasks because participants had to compute the global form of the object to extract the global shape, orientation, or size from objects made of parts having different shapes, orientations, or sizes. This analysis of global form was not necessary when matching for color or luminance. Because the objects were not made of parts having different colors or levels of luminance, attention to a single local part was sufficient to perform the task. It is possible that attention to local parts of objects may be sufficient to prevent semantic processing. This study was designed to dissociate these three explanations, to evaluate their respective validity, and to elucidate further the mechanisms responsible for semantic interference effects in matching based on the physical properties of pictures. General Method Unless otherwise mentioned, the same method was followed in all the reported experiments.
Participants The participants were 90 undergraduate students in psychology from the Universite Rene Descartes Paris V who participated for course credit. All had normal or corrected-to-normal vision. They were naive about the purpose of the experiment. There were 15 participants in each of the six reported experiments. Stimuli A set of 16 outline drawings of objects was used. There were eight animals (rabbit, mouse, hen, bear, fox, fish, tortoise, and crocodile) and eight vehicles (car, lorry, tractor, sailboat, helicopter, train, plane, and tank). The 16 pictures are illustrated in Figure 1. At a viewing distance of 57 cm, the mean angular size of the stimuli was 2.12° vertically (ranging from 1.6° to 2.7°) and 3.06° horizontally (ranging from 2.6° to 3.5°). Apparatus The stimuli were displayed on a black-and-white videomonitor. They were generated through a Hewlett-Packard (Vectra RS/20C) microcomputer equipped with a VGA graphic card. The screen resolution was 640 X 480 pixels (VGAhi). The stimuli were presented in black (VGA Color Number 1) on a light gray background (VGA Color Number 10). In a dark room the luminance of the outline drawings was 0.02 cd/m2, and the luminance of the background was 31.1 cd/m2. The contrast (Michelson) was 99.8%. Two Morse keys were used for response. Procedure The sequence of events on a trial was as follows: a fixation point (square 4 X 4 pixels subtending 0.037° of visual angle) was displayed centraHy for 500 ms. It was followed 500 ms later by a reference picture. The reference stimulus was centrally presented
Figure 1. The set of 16 outline drawings of objects used in the experiments. for 150 ms (9 frames at a refreshing rate of 16.7 ms). After a delay of 500 ms, during which the fixation point was represented, a pair of pictures was displayed for 150 ms left and right of fixation. The center of the lateral object was located 3° from fixation. According to the experiment, participants were asked to decide whether the right or the left picture had the same direction of motion as the reference picture (moving up or down in Experiments la and Ib); the same texture as the reference picture (density of dots or line orientation in Experiments 2a, 2b, and 2c); or the same "dominant" color as the reference picture (two thirds of the outline contour was red, and one third was green or the reverse in Experiment 3). Participants gave their response by pressing the left or the right key according to the spatial location of the target. The onset of the lateral stimuli activated the clock of the computer, which was stopped when the participant pressed a response key. The paradigm is presented in Figure 2 for a matching task based on motion. Before the experiment, participants were shown an example of the stimuli to be matched and were given 20 practice trials on the task. Errors were indicated by the message "Erreur" centrally displayed for 300 ms immediately after the execution of the response. These trials were not replaced. The intertrial interval was fixed at 2,000 ms after the execution of the response. The practice session was followed by a single block of 192 trials. The entire experiment lasted about 20 min. Design There were six matching conditions determined by the relations (a) between the reference picture and the target (same direction of motion, same texture, or same dominant color as the reference, according to the experiment) and (b) between the reference picture and the distractor (different direction of motion, texture, or dominant color). The reference stimulus and the target were either (a) physically identical, (b) physically different and semantically re-
ATTENTION TO GLOBAL INFORMATION REFERENCE
spatial location of the target; the semantic category of the reference object; the direction of motion (or, in Experiments 2 and 3, the texture or the color); and the target and distractor conditions were equally and randomly represented.
t
Experiment 1: Direction of Motion
1 TARGET
587
DISTRACTOR
Figure 2. Illustration of the paradigm used in the present study. A reference outline drawing of object was centrally displayed for 150 ms and followed, 500 ms later, by two objects (a target and a distractor) presented 3° left and right of fixation for 150 ms. The arrows illustrate the direction of motion used in Experiments la and Ib.
lated (two animals or two vehicles), or (c) semantically unrelated (an animal and a vehicle). For each of these three target conditions, there were two distractor conditions. The distractor and the reference picture were either semantically related or semantically unrelated. The distractor was never physically identical to the reference object. Depending on the experiment, participants were asked to respond to the direction of motion, to the texture, or to the color regardless of the object on a trial. For each of the six experimental conditions, each of the 16 reference objects was presented once with the target located on the left and once with the target located on the right, thus yielding 32 trials per condition. For physically identical stimuli, the 16 different reference objects were associated with the same objects presented once on the right and once on the left of fixation. For the semantically related pictures, each of the eight objects from the two semantic categories was randomly associated with one of the seven objects belonging to the same semantic category. The pairs of semantically related reference and target stimuli were randomly chosen to generate 16 different pairs with the target located on the left and the same 16 pairs with the target located on the right of fixation. The only constraint for the choice of the pairs was that the reference and target objects had approximately the same angular size vertically and horizontally, to avoid any bias toward form similarity in the selection of the target. For semantically unrelated items, each of the eight reference objects from the two semantic categories was associated with one of the eight items of the other semantic category. Similarly to semantically related stimuli, the choice of the reference and target stimuli was made on the basis of similarity in global shape (i.e., items having a similar angular size vertically and horizontally). Sixteen pairs were presented with the target located left of fixation, and the same pairs were used with the target located on the right. The same procedure as that described for the target conditions was used to choose the semantically related or semantically unrelated distractors. Again, stimuli having a similar global shape to the target and reference stimuli were used to build the pairs. The
Anatomical and physiological evidence indicates that separate channels exist for form (orientation and size), color, and motion information (see Desimone & Ungerleider, 1989; Livingstone & Hubel, 1987; Maunsell & Newsome, 1987; Zeki, 1990a, for reviews). The utility of this account as an explanation for the different results with form and color-matching tasks (Boucart & Humphreys, 1994) was tested in Experiment 1 with a task requiring matches to motion information. It is well known that some regions in the visual cortex are specialized for motion processing. For instance, the middle temporal area (MT) exhibits systematic selectivity for movement (Albright, Desimone, & Gross, 1984; Livingstone & Hubel, 1987; Maunsell & Newsome, 1987; Maunsell & van Essen, 1983). Although some physiological and anatomical studies show that the form, color, and motion pathways are not totally independent, because connections exist between area MT and area V4 (involved in the processing of form and color; Desimone & Ungerleider, 1989; Zeki, 1990a), neuropsychological studies on brain-damaged patients support a dissociation between the processing of motion and of form and color. For instance, there are documented cases in the literature showing a selective loss of color vision usually following lesions in the lingual and fusiform gyri (Damasio, Yamata, Damasio, Corbett, & McKee, 1980; Pearlman, Birch, & Meadows, 1979; see Zeki, 1990a, 1990b, for reviews), along with good motion perception (Humphreys, Donnelly, & Riddoch, 1993) and good form perception (Heywood, Wilson, & Cowey, 1987). In contrast, there are recorded cases of neurologically impaired patients in which severe disturbances in face and object perception were not accompanied by achromatopsia (Hecaen, Angelergues, Bernhard, & Chiarelli, 1957; Michel, Perenin, & Sieroff, 1986) and also selective impairments of motion perception with intact processing of stationary stimuli and colors following lesions of occipitotemporal or occipitoparietal areas (Riddoch, 1917; Vaina, 1989; Zeki, 1991; Zilh, Von Cramon, & Mai, 1983; Zilh, von Cramon, Mai, & Schmid, 1991). Experiment 1 was designed to test whether motion information can be attended selectively without there being access to semantic information. Experiment la: Continuous Motion Method The reference stimulus appeared nine times for 16.7 ms with a shift of 2 pixels moving either up or down. The delay between each display was 16.7 ms, thus making the apparent motion continuous. The same procedure was used for the left and the right lateral objects. One stimulus moved up and the other moved down. Participants were required to decide whether the left or the right
588
M. BOUCART, G. HUMPHREYS, AND J. LORENCEAU
object moved in the same direction as the reference object by pressing the left or the right key. An example is shown in Figure 2.
/DU-
700-
•
DISTRACTOR SR
650-
—o—
UlSiiKACTORsu
Results RTs longer than 2,000 ms and shorter than 150 ms were discarded. Less than 1% of the data were eliminated this way. Analyses of variance (ANOVAs) were carried out on the correct RTs and on the error rates using both participants (Fj) and reference stimuli (F2) as random factors. There were five within-subject factors: the spatial location of the target (left vs. right); the semantic category of the reference object (animal vs. vehicle); the direction of motion (up vs. down); the target condition (physically identical, semantically related, or semantically unrelated); and the distractor condition (semantically related vs. unrelated). The mean RT was 585 ms, and the mean error rate was 4.2%. There was no significant difference in RTs between left and right targets (587 vs. 583 ms), but right targets were matched more accurately than left targets (3% vs. 5.4%), Fj(l, 14) = 7.3, p < .016, and F2(l, 15) = 5.5, p < .03. Vehicles were matched faster than animals (578 vs. 592 ms), Fj(l, 14) = 7.9, p < .013, and F2(l, 15) = 5.45, p < .032, but animals were matched more accurately than vehicles (3.6% vs. 4.9%), Fj(l, 14) = 3.1, p < .09, and F? < 1. There was no significant main effect of the direction of motion (up, 589 ms and 4.1% vs. down, 580 ms and 4.4%). These factors did not significantly interact with the other experimental variables. The results displayed in Figure 3 are averaged over target location, semantic category, and direction of motion. There was a significant main effect of target condition both for RTs, F&, 28) = 7.76, p < .002, and F2(2, 30) = 16.7, p < .001, and for errors, Fx(2, 28) = 3.1, p < .059, and F2(2, 30) = 5.9, p < .007. The shorter RTs and the lower error rate were observed for physically identical items (564 ms and 2.7%), followed by semantically related pictures (590 ms and 4.1%). The difference between these two target conditions was significant for RTs, F1(l, 14) = 7.4, p < .015, and F2(l, 15) = 22.2, p < .001, but not for errors, F^l, 14) = 1.2, ns, and F2 < 1. Performance for semantically related items (without physically identical pictures) and semantically unrelated pictures (600 ms and 6.4%) differed significantly for errors only, Fj(l, 14) = 6.6, p < .021, and F2(l, 15) = 7.99, p < .012. RTs were longer and the error rate was higher when the distractor was semantically related to the reference stimulus (592 ms) than when it was unrelated (577 ms), Fj(l, 14) = 8.5, p < .01, and F2(l, 15) = 5,p< .038. The error rate was equivalent for the two distractor conditions (related, 4.4% vs. unrelated, 4%), F^l, 14) = 1, ns, and F2 < 1. No significant interaction was found between target condition and distractor condition. Discussion The fact that effects of semantic relations between stimuli were confined to either RTs or errors suggests that semantic interference was less pronounced on the processing of mo-
600-
§ g ec8
•^—^ ^•••^"^L 1— —i I
r
550500-
A*n*
^ PI
SR
SU
Target Conditions
25'
B 20-
Z §
15-)
uj
^
ioH 5-
PI
SR
SU
Target Condition*
Figure 3. Mean correct response times (A) and mean error rates (B) for Experiment la as a function of the three target conditions (physical identity [PI], semantically related [SR], and semantically unrelated [SU]) and the two distractor conditions (SR and SU). The vertical bars represent standard errors.
tion information than it was on the processing of form information observed in previous studies. For instance, semantically related target and reference objects were matched faster (by 84 ms relative to 10 ms in the present experiment) and more accurately (by 8.5% relative to 2.3% here) than semantically unrelated items in a task requiring orientation matching (Boucart & Humphreys, 1994, Experiment Ib); similarly, longer RTs (by 33 ms relative to 15 ms here) and a higher error rate (by 4.1% relative to 0.4% here) were found for semantically related distractor and reference objects relative to unrelated stimuli in a size matching task (Boucart & Humphreys, 1994, Experiment 23).1 1
The experiments of Boucart and Humphreys (1994) and those reported here are comparable because the same paradigm and the
ATTENTION TO GLOBAL INFORMATION
Nevertheless, evidence for the processing of form information and for access to stored object knowledge was apparent. There was a substantial advantage for physically identical stimuli, a lower error rate for semantically related target and reference objects relative to unrelated items, and longer RTs for semantically related distractors, suggesting that motion information was not attended without form processing and object identification being engaged. These results can be accounted for in several ways. A first possibility concerns the presentation conditions. Stimuli were displayed 16.7 ms before they started to move. This earlier exposure of stationary stimuli gave participants an opportunity to identify the objects from their form prior to movement taking place. Also, with practice, participants could notice that when the lateral object was physically identical to the reference object it always moved in the same direction. Therefore, a decision could be based simply on form information. A second possibility concerns the correspondence problem in apparent motion (i.e., the process by which an object maintains its identity when seen in different spatial locations at different times; Green, 1986; Ullman, 1980). Several studies have been designed to determine which stimulus parameters contribute to the correspondence process. These experiments typically use a procedure in which either a centrally presented stimulus is followed by two flanking figures (Ramachandran, Ginsburg, & Anstis, 1983; Ullman, 1980) or a sequence of frames is presented with several elements arranged on a circular ring (Green, 1986; Shechter, Hochstein, & Hillman, 1988). In both procedures, participants have to report the direction of apparent motion. The results show that motion toward the neighbor having the same shape (Shechter et al., 1988), the same orientation (Ullman, 1980), or the same spatial frequency content (Green, 1986; Ramachandran et al., 1983) is preferred, suggesting that physical identity or physical similarity is used by the apparent motion correspondence process (although Navon, 1976, observed no effect of physical similarity in the perception of apparent motion with English and Hebrew letters). This effect has been found to be robust over variations in temporal intervals and the spatial separation of the stimuli (Green, 1986; Ramachandran et al., 1983). Interpretations refer either to a "long range" mechanism (Braddick, 1974) involving identification of the form and the tracking of its position over time (Anstis, 1978; Cavanagh & Mather, 1985) or to the computation of a "preference metric" (Ullman, 1980) calculated by determining the number of feature matches between an object in frame n and those in frame n + 1 (Green, 1986). same presentation conditions were used. The experiments differed only on the attended physical property. The pictures were not all the same. For instance, in the orientation matching task (Boucart & Humphreys, 1994 Experiment Ib), the main axis was oblique for half of the stimuli and horizontal for the other half. In the present experiments, the global orientation of the stimuli was as homogeneous as possible to avoid a bias toward global orientation in the matching tasks. Note, however, that all the main results generalized across items.
589
The advantage for physically identical stimuli may be accounted for by the correspondence process. Moreover, because semantic similarity is correlated with physical similarity (Mervis & Rosch, 1981), the interference from semantically related distractors and the lower error rate for semantically related target and reference objects, relative to semantically unrelated objects, are also compatible with the view of correspondence in processing apparent motion. This explanation is consistent with participants' reports and with recent data showing that the advantage for physically identical stimuli decreases with the increase in interstimulus interval (ISI) between stimuli. This study (Boucart & Humphreys, 1995) involved a matching task based on shape in which the ISI between the reference picture and the lateral pictures was manipulated. Seven ISIs were used: 17, 33, 50, 100, 500,1,500, and 2,000 ms. Shorter ISIs (lower than 500 ms) produced both strong apparent motion and a large advantage for physically identical items, and both effects decreased as the ISI increased. These different explanations are not mutually exclusive. Each factor might have played a role in the effects of form and semantic information on judgments of motion. In Experiment Ib we tested the first hypothesis, that is, whether the advantage for physically identical stimuli and the interference from semantically related distractors was due to preexposure to stationary forms at the start of each trial. Here we used a longer exposure time for stationary stimuli than in Experiment la. If semantic interference and form priming effects came about from the preexposure of static primes, then both effects should be enhanced, because there will be relatively more time to identify preexposed stimuli.
2 The dissociation between the effects of form information, as shown in the physically identical condition, and effects of semantic information found with physically different stimuli, is based on the result of a previous study (Boucart & Humphreys, 1992). In this study we compared performance for structurally similar meaningful and meaningless fragmented forms. The two types of stimuli were similar in terms of their global shape and the spatial location of the fragments, the only difference being that the alignment of fragments was reduced in the meaningless forms. This procedure affected their identifiability. We observed an effect of semantic relations on matching between meaningful fragmented forms but not between meaningless versions of the forms or between inverted meaningful pictures. However, in both cases matching was facilitated when the reference stimulus and the matching target were physically identical relative to when they were physically different. Thus the elimination of semantic interference for meaningless forms and for inverted meaningful forms did not affect the advantage for physically identical items, suggesting that matching was mainly based on presemantic form information. Note that an advantage for physically identical items relative to physically different and semantically related items has been reported in other studies (e.g., Bartram, 1976, with pictures, and Posner & Mitchell, 1967, with uppercase and lowercase letters).
590
M. BOUCART, G. HUMPHREYS, AND J. LORENCEAU
Experiment Ib: Motion With a Long Display Time
/W
Method
700-
£
The stimuli and the procedure were exactly the same as for Experiment la, except that the stimuli were centrally displayed for 68 ms (4 frames) and presented again 16.7 ms later 18 pixels higher (or lower) for 68 ms.
650-
d>
1
600-
i!
§ g-n
Results Like Experiment la, RTs longer than 2,000 ms and shorter than 150 ms were discarded. ANOVAs were carried out on the correct RTs and on the error rates using both participants (Fx) and reference stimuli (F2) as random factors. The factors were the same as those in Experiment la. The mean RT was 635 ms, and the mean error rate was 8.8%. RTs tended to be longer and the error rate higher when the target was located left of fixation (641 ms and 10.1%) than on the right (629 ms and 7.5%). The differences were not statistically significant, Fj(l, 14) = 1.7, ns, and F2(l, 15) = 2.3, ns, for RTs, and Fj(l, 14) = 3.04, p < .09, and F2(l, 15) = 3.5, p < .078, for errors. Vehicles tended to be matched more accurately than animals (7.7% vs. 9.9%); F^l, 14) = 3.25, p < .089, and F2(l, 15) = 2.5, ns, but RTs were equivalent for the two semantic categories (animals, 634 ms vs. vehicles, 636 ms). Pictures moving down were matched faster and more accurately than pictures moving up (down, 618 ms and 7.6% vs. up, 652 ms and 10.1%), F^l, 14) = 19, p < .001, and F2(l, 15) = 2.54, ns, for RTs, and F^l, 14) = 5.28, p < .038, and F2(l, 15) = 3.4, p < .08 for errors. These factors did not interact significantly with the other experimental variables. The results displayed in Figure 4 are averaged over spatial location, semantic category, and direction of motion. The main effect of target condition was significant for RTs, Fj(2, 28) = 14, p < .001, and F2(2, 30) = 8.9, p < .001, but not for errors, Fj(2, 28) = 3.1, p < .058, and F2(2, 30) = 2.7, ns. RTs and errors increased from physically identical items (616 ms and 7%) to semantically related stimuli (639 ms and 9.4%), Fj(l, 14) = 8.6, p < .01, and F2(l, 15) = 8.2, p< .01, for RTs, and F^l, 14) = 4.4, p < .052, and F2(l, 15) = 2.1, ns, for errors. RTs and errors did not differ significantly between semantically related and semantically unrelated pictures (related, 639 ms and 9.4% vs. unrelated, 649 ms and 10.1%), Fj(l, 14) = 1.1, ns, and F2(l, 15) = 2.36, ns, for RTs and both Fs < 1 for errors. RTs were longer and the error rate higher when the distractor was semantically related to the reference picture than when it was unrelated (related, 645 ms and 9.6% vs. unrelated, 624 ms and 8%), F^l, 14) = 18.8, p < .001, and F2(l, 15) = 12.3, p < .003, for RTs, and F^l, 14) = 5.3, p < .034, and F2(l, 15) = 3.55, p < .076, for errors. The comparison of continuous motion (Experiment la) and shift in position (Experiment Ib) showed a significant main effect of experiment, with longer RTs and a higher
T
^5*
oc
T
T^~--—* o 1 1
•«
I
550500-
PI
•
DISTRACTOR SR
— o—
DISTRACTOR SU
SR
SU
Target Conditions
25'
B 20-
»
o
1510-
5-
PI
SR
SU
Target Condition*
Figure 4. Mean correct response times (A) and mean error rates (B) for Experiment Ib as a function of the three target conditions (physical identity [PI], semantically related [SR], and semantically unrelated [SU]) and the two distractor conditions (SR and SU). The vertical bars represent standard errors.
error rate for Experiment Ib (635 ms and 8.8%) than for Experiment la (585 ms and 4.2%), F^l, 28) = 5.34, p < .027, and F2(l, 30) = 87.8, p < .001, for RTs, and Fx(l, 28) = 6.05, p < .019, and F2(l, 30) = 31.4, p < .001, for errors. The main effect of target condition was significant both for RTs, FJ2, 56) = 19.4, p < .001, and F2(2, 60) = 19.8, p < .001, and for errors, F^2, 56) = 7.46, p < .001, and F2(2, 60) = 7.44, p < .001. This effect was due to an advantage for the physical identity condition in the two experiments. The main effect of distractor condition was also significant both for RTs, F^l, 28) = 26, p < .001, and F2(l, 30) = 17.06, p < .001, and for errors, F^l, 28) = 5.79, p < .022, and F2(l, 30) = 3.9, p < .053. No significant interaction involving experiment was observed.
ATTENTION TO GLOBAL INFORMATION
Discussion The results of Experiment Ib are very similar to those observed with continuous motion in Experiment la. There was again an advantage for physically identical target and reference objects relative to physically different items. Also, interference from semantically related distractors was observed, on both RTs and errors. The main difference between the two experiments was that RTs were longer and the error rate higher in Experiment Ib than in Experiment la. The difference in RTs and errors between the studies can be attributed to the contrasting stimulus exposure time before motion onset. Motion started after a stationary stimulus had been displayed for 16.7 ms in Experiment la, and it started after 68 ms in Experiment Ib. The difference in RTs between the two experiments corresponds exactly to the difference in time before the beginning of motion (51 ms). The overall RT difference was also equivalent across the three target conditions. The higher error rate in Experiment Ib can be explained by the fact that stimuli moved once in Experiment Ib, whereas they moved continuously in Experiment la, thus making the discrimination of the direction of motion easier in Experiment la. The advantage for physically identical stimuli and the interference from semantically related distractors were not larger in Experiment Ib than in Experiment la, despite the longer exposure of stationary stimuli in Experiment Ib. This result suggests either that the identification process can be engaged with a 16.7 ms display time and that a longer exposure time does not add anything once this processing has started, or that motion information could not be attended selectively without form and object processing taking place. Nevertheless, interference from semantic information was less pronounced than when participants matched on aspects of form information (Boucart & Humphreys, 1992, 1994). This difference can be explained either by motion information being less efficient than characteristics of form in activating stored object representations (Lorenceau & Boucart, in press, found no difference in performance between meaningful and meaningless forms in a discrimination task in which participants were asked to decide whether a stimulus moved clockwise or counterclockwise), by the fact that the output of motion processing was available before the output of semantic processing for decision, or by both factors. This point is developed in the General Discussion. In Experiment 2 we tested whether semantic interference also occurs when participants are asked to attend to surface information. Studies investigating the role of surface versus edgedbased descriptions in object recognition suggest that surface information is less efficient than form information in activating stored semantic representations (Biederman & Ju, 1988; Ostergaard & Davidoff, 1985; Price & Humphreys, 1989). Semantic interference within a task requiring the perception of surface information was not directly tested in Boucart and Humphreys (1994), because objects comprised only colored outlines. In Experiment 2 we tested whether
591
semantic interference occurred in a matching task requiring comparisons of the internal surface of objects. We chose a matching task on the basis of the density of dots filling the objects displayed.3 Failure to find semantic effects on the matching of such stimuli would suggest that semantic information can be filtered out when participants attend to surface properties.
Experiment 2: Texture Discrimination
Experiment 2a: Texture Discrimination on the Density of Dots Method The outline drawings of objects were filled with dots made of 1 pixel. There were two densities: the spacing between the dots was either 2 pixels (high density) or 4 pixels (low density). An example is shown on Figure 5. Participants were asked to indicate whether the left or the right picture had the same density of dots as the reference object by pressing the left or the right key.
Results As in the previous experiments, RTs longer than 2,000 ms and shorter than 150 ms were discarded. Less than 1% of the data were eliminated this way. ANOVAs were carried out on the correct RTs and on the error rates using both participants (Fj) and reference stimuli (F2) as random factors. There were five within-subject factors: the spatial location of the target (left vs. right); the semantic category of the reference object (animal vs. vehicle); the density of dots (high vs. low); the target condition (physically identical, semantically related, or semantically unrelated); and the distractor condition (semantically related vs. unrelated). The mean RT was 558 ms, and the mean error rate was 12.9%. Performance was equivalent for left targets (558 ms and 12.4%) and for right targets (558 ms and 13.5%). Animals were matched faster and more accurately than vehicles (546 ms and 10.5% vs. 568 ms and 15.3%), Fj(l, 14) = 11.07, p < .004, and F2(l, 15) = 4,p < .06, for RTs and F^l, 14) = 8.6, p < .01, and F2(l, 15) = 3.5, p < .075 for errors. Stimuli with low density of dots were matched faster and more accurately than were stimuli with high density (546 ms and 10.4% vs. 568 ms and 15.4%), Ft(l, 14) = 5.6, p < .03, and F2(l, 15) = 6.5, p < .022, for RTs, and F^l, 14) = 3.9, p < .65, and F2(l, 15) = 2.57, ns, for errors. The results are graphically presented in Figure 6. 3 In previous studies (Boucart & Humphreys, 1992,1994) physical properties that do not correspond to the physical characteristics of real objects were also used. For instance, a rabbit was arbitrarily drawn with a "round" shape (Boucart & Humphreys, 1992), or a lorry was displayed with a smaller size than a mouse (Boucart & Humphreys, 1994). Semantic effects were found even with these "nonnatural" conditions. It follows that nonnatural surface information should not be critical for the absence of semantic effects on texture.
592
M. BOUCART, G. HUMPHREYS, AND J. LORENCEAU REFERENCE
TARGET
DISTRACTOR
Figure 5. An example of the two textures varying in dot density used in Experiment 2a: low dot density (the dots were separated by 4 pixels) and high dot density (the dots were separated by 2 pixels). Participants were asked to decide which of the two lateral figures had the same density of dots as the reference object.
The main effect of target condition was significant both for RTs, Fr(2, 28) = 13.9, p < .001, and F2(2, 30) = 8.1, p < .001, and for errors, Fx(2, 28) = 14.5, p < .001, and F2(2, 30) = 9.59, p < .001. RTs were faster and the error rate was lower for physically identical items (534 ms and 8.9%) than for semantically related pictures (567 ms and 13.8%), Fj(l, 14) = 12.05, p < .0037, and F2(l, 15) = 16.81, p < .001, for RTs, and Fj(l, 14) = 30.6, p < .001, and F2(l, 15) = 11.78, p < .004, for errors. RTs and errors increased from semantically related to semantically unrelated items (574 ms and 16%), but the differences failed to reach statistical significance, Fj(l, 14) = 2, ns, and F2 < 1 for RTs, and Fj(l, 14) = 2.23, ns, and F2(l, 15) = 1.97, ns, for errors. The mean RT was longer and the mean error rate higher when the distractor was semantically related to the reference stimulus (566 ms and 13.4%) than when it was semantically unrelated (550 ms and 12.6%). The difference was significant only for RTs with participants as a random factor, Fj(l, 14) = 5.4, p < .034, and F2(l, 15) = 1.5, ns. There was a significant interaction between target condition and texture for errors, Fj(2, 28) = 6.2, p < .006, and F2(2, 30) = 5.9, p < .007. The error rate was higher for stimuli with a high density of dots than for those with a low density in all three target conditions, although the difference was more pronounced for semantically unrelated stimuli (high density, 21.2% vs. low density, 10.9%), F^l, 14) = 8.2, p < .012, and F2(l, 15) = 8.52, p < .01, than for physically identical stimuli (high density, 10.2% vs. low density, 7.7%, both Fs < 1) and semantically related pictures (high density, 15% vs. low density, 12.7%, both Fs < 1).
Discussion The lack of facilitation for semantically related target and reference stimuli, relative to unrelated stimuli, and the fact that distractor interference was confined to RTs (and reliable only across participants) suggests that semantic processing was less pronounced here than when participants matched on characteristics of form, such as global shape,
orientation, or size (Boucart & Humphreys, 1992, 1994). Nevertheless, unlike matching based on color or luminance (Boucart & Humphreys, 1994), evidence for form processing was not eliminated because there was an advantage for physically identical items. In contrast to previous tasks requiring the matching of color, luminance, or polarity of contrast, where the outline contour of the object was composed of a single color, the surface characteristics of stimuli here were not homogeneous. The density of the dots varied locally in different parts of the objects since, because of the constraints of filling, small parts of objects could not contain many dots (see Figures 5 and 6). This may have led participants to focus attention on parts of the lateral stimuli where the difference in density was the most discriminable. Attention to local parts may be crucial in reducing access to stored object representations. This locally based processing contrasts with the global shape, orientation, size, and even the motion matching tasks we have examined previously (Boucart & Humphreys, 1992, 1994, Experiments la and Ib here), where attention may be given to global form information. We return to this point in the General Discussion. The better performance for stimuli with a low density of dots suggests that participants did not match on the basis of contrast information. We have previously shown that matching based on contrast is not affected by form or semantic relations between the stimuli (Boucart & Humphreys, 1994). If matches were based on contrast here, it would explain why only a small semantic effect occurred. However, for matches based on contrast, darker stimuli should be matched faster or more accurately than lighter stimuli, because the darker stimuli show a greater change relative to the background. We found the opposite result. Another possible explanation for the small semantic interference effect could be that the dots masked the internal contours of the object's parts and hence impaired object identification. This masking would in effect make the outline drawing like silhouettes (see Figure 5). There is evidence that silhouettes are more difficult to identify than outline drawings of objects (Riddoch & Humphreys, 1987), and no semantic interference was found in a matching task on size with silhouettes as stimuli in a previous study (Boucart & Humphreys, 1994, Experiment 2b). However, if dots impaired the processing of contour information, and hence object identification, then semantic interference should have been larger for stimuli having a low dot density because outline contours were less masked in this condition. Examination of Figure 6 shows that, if anything, semantic relations between stimuli affected performance more for stimuli with a high dot density. Nevertheless, a control experiment was designed to test the masking account with stimuli having a thick contour. If object identification was prevented by contour masking in Experiment 2a, then stronger evidence for object identification (e.g., stronger semantic interference from distractors) should emerge here.
593
ATTENTION TO GLOBAL INFORMATION
Results
Experiment 2b: Texture Discrimination on Stimuli With Thick Contour Method The procedure was the same as in Experiment 2a except that the width of outline contour of objects was 2 pixels. An example trial is presented in Figure 7.
As in previous experiments, RTs longer than 2,000 ms and shorter than 150 ms were discarded. Less than 1% of the data were eliminated this way. ANOVAs variance were carried out on the correct RTs and on the error rates using both participants (Fa) and reference stimuli (F^ as random factors. The factors were the same as those of Experiment 2a.
Low Density
25
750' 700-
•
DISTRACTOR SR
—-o-—
DISTRACTOR SU
20-
65015-
s
600-
UJ
ID-
550500-
450? PI
SR
SU
PI
Target Condition*
SR
SU
Target Conditions
High Density 25
750
DISTRACTOR SR 700
20-
DISTRACTOR SU
i P | |
600
Ul
a*
10
550
5-
500
PI
SR
Target Conditions
SU
PI
SR
SU
Target Conditions
Figure 6. Mean correct response times and mean error rates for Experiment 2a for textures made of low-density and high-density dots as a function of the three target conditions (physical identity [PI], semantically related [SR], and semantically unrelated [SU]) and the two distractor conditions (SR and SU). The vertical bars represent standard errors.
594
M. BOUCART, G. HUMPHREYS, AND J. LORENCEAU
The mean RT was 603 ms, and the mean error rate was 13.4%. Right targets tended to be matched faster and more accurately than left targets (595 ms and 12% vs. 608 ms and 14.7%), Fj(l, 14) = 3.03, ns, and F2 < I for RTs and Fl and F2 < 1 for errors. RTs were shorter and the error rate was lower for animals than for vehicles (593 ms and 11.8% vs. 610 ms and 14.6%), F^l, 14) = 4.49, p < .05, and F2 < 1 for RTs, and Fl and F2 < 1 for errors. Stimuli with a low dot density were matched faster and more accurately than were stimuli with a high dot density (594 ms and 10.8% vs. 609 ms and 15.7%), F^l, 14) = 1.45, ns, and F2(l, 15) = 3.67, p < .07, for RTs, and Fx(l, 14) = 8.07, p < .012, and F2(l, 15) = 2.6, ns, for errors. The results are presented graphically in Figure 8. The main effect of target condition was significant both for RTs, Fx(2, 28) = 10.82, p < .001, and F2(2, 30) = 27.62, p < .001, and errors, F-fe, 28) = 7.68, p < .002, and F2(2, 30) = 13.32, p < .001. Physically identical stimuli were matched faster and more accurately than semantically similar objects (564 ms and 8.3% vs. 618 ms and 16.4%), F^l, 14) = 12.2, p < .0036, and F2(l, 15) = 35.8, p < .001, for RTs and Fj(l, 14) = 11.85, p < .0039, and F2(l, 15) = 31.65, p < .001, for errors. The difference in RTs and in error rate between semantically related pictures and semantically unrelated items (625 ms and 15.3%) was not significant. RTs tended to be longer and the error rate higher for semantically related distractors and reference objects than for semantically unrelated items (607 ms and 14.1% vs. 598 ms and 12.6%), but the difference was significant neither for RTs (both Fs < 1) nor errors, Fj(l, 14) = 2.35, ns, and F2(l, 15) = 2.98, ns.
750
•—
DISTRACTOR SR
—-0-—
DISTRACTOR SU
7006500>
600:
|
55
°"
500-
450^ PI
The results show that the interference effect from semantically related distractors observed in Experiment 2a disappeared with stimuli having a thick contour. This result suggests that the small effect of semantic information in Experiment 2a was not due to difficulties in identifying the stimuli, given that there was no effect of semantic relations REFERENCE
TARGET
Figure 7. An example of the stimuli used in Experiment 2b. The densities of dots are the same as those used in Experiment 2a, but the outline contours of the objects are thicker in Experiment 2b than in Experiment 2a. Participants were asked to decide which of the two lateral figures had the same density of dots as the reference object.
SU
Target Conditions
25'
B
20-
o
105-
PI
Discussion
SR
SR
SU
Target Conditions
Figure 8. Mean correct response times (A) and mean error rates (B) for Experiment 2b for textures made of low-density and high-density dots as a function of the three target conditions (physical identity [PI], semantically related [SR], and semantically unrelated [SU]) and the two distractor conditions (SR and SU). The vertical bars represent standard errors.
between stimuli with a more discriminable contour. Moreover, the longer RTs (45 ms) for stimuli having a thick relative to a thin contour (Experiment 2a) suggests that thick contours impaired the discrimination of texture by masking surface information. Like Experiment 2a, the tendency for longer RTs and a higher error rate for stimuli having a high, relative to a low, dot density suggests that participants did not base their decision on contrast information, given that stimuli with a high dot density have more contrast (and more energy) than those having a low dot density (see Figures 5 and 7). In Experiment 2c we tested the generality of Experiment 2b by having participants match objects with different internal textures: horizontal and vertical lines rather than dots. The use of oriented line textures is of interest here because
595
ATTENTION TO GLOBAL INFORMATION
Boucart and Humphreys (1994) found strong evidence for object identification with matching based on the global orientation of the pictures. If what was crucial to those results was the processing of global form information, semantic interference should be reduced when a local orientation judgment is required.
Experiment 2c: Texture With Horizontal Versus Vertical Bars Method The 16 outline drawings of objects were filled with either horizontal or vertical bars having a width of 2 pixels. The space between bars was 4 pixels. An example is shown in Figure 9. Participants were required to decide whether the left or the right objects matched the reference object in terms of the orientation of the bars. Participants responded by pressing the left or the right key.
and the other variables. The data are presented in Figure 10, averaged over target location, semantic category, and texture. The main effect of target condition was significant both for RTs, Fj(2, 28) = 18.74, p < .001, and F2(2, 30) = 43.62, p < .001, and for errors, Fj(2, 28) = 13.1, p < .001, and F2(2, 30) = 11.53, p < .001, when the condition of physical identity was included in the analysis. RTs were shorter and the error rate lower for physically identical items (596 ms and 4.4%) than for semantically related pictures (664 ms and 11.2%), F^l, 14) = 22.2, p < .001, and F2(l, 15) = 40.03, p < .001, for RTs, and F^l, 14) = 16.4, p < .001, and F2(l, 15) = 19.56, p < .001, for errors. 750
700-
•
Results
650-
600-
As in previous experiments, RTs longer than 2,000 ms and shorter than 150 ms were discarded. Less than 1% of the data were eliminated this way. ANOVAs were carried out on the correct RTs and on the error rates using both participants (Fj) and reference stimuli (F2) as random factors. The factors were the same as those of Experiments 2a and 2b. The mean RT was 647 ms, and the mean error was 8.8%. For an equivalent error rate, targets tended to be matched faster when they were located on the left than when located on the right of fixation (635 ms and 8.9% vs. 658 ms and 8.7%), F^l, 14) = 1.8 ns, and F2 < 1. There was no significant difference in RTs and in errors between animals (643 ms and 9.2%) and vehicles (650 ms and 8.1%), Ft and F2 < 1. There was no difference in RTs between textures composed of vertical bars (643 ms) and textures composed of horizontal bars (650 ms), but accuracy was better for texture composed of vertical bars (6.7% vs. 10.6%), Fj(l, 14) = 6.37, p < .023, and F2(l, 15) = 10.02, p < .006. There was no significant interaction between these factors
a
550
' •
ac
-
DISTRACTOR SR
500-
45Q0 PI
SR
SU
Target Condition!
25'
B 20-
Ul
*
10
REFERENCE
5-
PI
SR
SU
Target Condition* TARGET
DISTRACTOR
Figure 9. An example of the stimuli used in Experiment 2c. The texture varies on the orientation of the line segments: horizontal versus vertical. Participants were asked to decide which of the two lateral figures had the same orientation of the line segments as the reference object.
Figure 10. Mean correct response times (A) and mean error rates (B) for Experiment 2c averaged over the two texture conditions (vertical vs. horizontal line segments) as a function of the three target conditions (physical identity [PI], semantically related [SR], and semantically unrelated [SU]) and the two distractor conditions (SR and SU). The vertical bars represent standard errors.
596
M. BOUCART, G. HUMPHREYS, AND J. LORENCEAU
No significant difference was found in error rate between semantically related (SR) and semantically unrelated (SU) items (SR, 11.2% vs. SU, 10.6%), but RTs tended to be longer for related than for unrelated stimuli (681 ms vs. 664 ms), Fj(l, 14) = 2.7, ns, and F2(l, 15) = 2.57, ns. RTs tended to be longer and the error rate higher when the distractor was semantically related to the reference stimulus (654 ms and 9.4%) than when the two items were semantically unrelated (640 ms and 8.2%), Fj,(l, 14) = 3.7, p < .07, and F2(l, 15) = 3.86, p < .065, for RTs, and F^l, 14) = 2.1, ns, and F2(l, 15) = 1.2, ns, for errors. No interaction between target condition and distractor condition was found. The comparison of the three experiments involving matching on texture information showed a significant effect of experiment on RTs, Fj(2,42) = 2.65, p < .085, and F2(2, 45) = 104.6, p < .001, and for errors, Fj(2, 42) = 3.8, p < .03, and F2(2, 45) = 3.12, p < .053. The effect of target condition was significant when physically identical items were included, Fx(2, 84) = 18, p < .001, and F2(2, 90) = 62.8, p < .001, for RTs, and Fj(2, 84) = 29.72, p < .001, and F2(2, 90) = 7.09, p < .001, for errors, but not when they were excluded, F1 < 1 and F2(l, 45) = 1.44, ns, for RTs, and both Fs < 1 for errors. The effect of distractor condition was significant with pictures as random factor for RTs, Fi(l, 42) = 2.77, ns, and F2(l, 45) = 7.05, p < .01, and with participant as random factor for errors, Fj(l, 42) = 5.45, p < .025, and F2(l, 45) = 0.12, ns. No significant interaction was found between experiments and target condition (both Fs < 2 for RTs and errors) or between experiment and distractor condition (both Fs < 1 for RTs and errors). Discussion The results of Experiment 2c show that, although effects due to semantic relations between stimuli were reduced when matching involved textures made of oriented line segments, a tendency for semantic interference was still present, and a robust advantage for physically identical stimuli remained. In these aspects, the present results, like those of Experiments 2a and 2b, differ from those observed when matching requires judgments of polarity of contrast, luminance, or color, where there is both no advantage for physically identical stimuli and not a trend for an effect of semantic information (Boucart & Humphreys, 1994). Also, relative to matching based on global orientation of the pictures (Boucart & Humphreys, 1994, Experiment Ib), attention to local oriented bars here reduced considerably the impact of semantic information, suggesting that the processing of the global configuration might be the determining factor mediating activation of stored object representations. In Experiment 3 we assessed directly whether the effects physical identity and semantic interference were determined by global form information being attended. We chose a physical dimension for which no semantic interference was observed in a previous study: color. In that study the outline
contour of the objects was drawn in a single color (red or green on a gray background), and attention to a small part was sufficient to perform the task. In Experiment 3 the objects were drawn in two colors: two thirds of the dots composing the outline contour of the picture were red and one third were green (or the reverse). Participants were required to decide which of the two lateral pictures had the same "dominant" color as the reference. The red and green dots were randomly distributed. This procedure ensured that participants could not base their judgment on a local part of the picture and that a processing of the global configuration was necessary to perform the task. If semantic interference depends on stimuli being processed as perceptual wholes, then effects of semantic relations between stimuli should be observed in the color-matching task here.
Experiment 3: Spatial Location of Color Information Method The procedure, the stimuli, and the task were the same as those used in previous experiments. The difference was that the stimuli were displayed with two colors, red and green, on a light gray background. For each picture, 67% of the dots were red (or green) and 33% had the other color. The red and green dots were randomly distributed for each picture and on each trial. Participants were asked to decide which of the two lateral objects had the same "dominant" color (more red or more green) as the reference picture.
Results As in the previous experiments, RTs longer than 2,000 ms and shorter than 150 ms were discarded. Less than 1% of the data were eliminated this way. ANOVAs were carried out on the correct RTs and on the error rates using both participants (Fj) and reference stimuli (F2) as random factors. There were five within-subject factors: the spatial location of the target (left vs. right); the semantic category of the reference object (animal vs. vehicle); the "dominant" color (more red or more green); the target condition (physically identical, semantically related, semantically unrelated); and the distractor condition (semantically related vs. unrelated). The mean RT was 607 ms, and the mean error rate was 2.4%. RTs were longer when the target was located on the right than on the left (618 vs. 597 ms), Fx(l, 14) = 14.3, p < .002, and F2(l, 15) = 4.1, p < .058. There was no significant difference in the error rate (right, 2.1% vs. left, 2.7%). No significant difference in performance was found between the two semantic categories (animals, 603 ms and 2.6% vs. vehicles, 612 ms and 2.2%, Fs < 1). There was no difference in performance between pictures whose dominant color was green (603 ms and 2.5%) and pictures whose dominant color was red (612 ms and 2.3%). These factors did not interact with the other experimental variables. The results presented in Figure 11 are averaged over target location, semantic category, and color. The main effect of target condition was significant both for RTs, Fj(2, 28) = 7.99, p < .001, and F2(2, 30) = 5.09,
597
ATTENTION TO GLOBAL INFORMATION
.087, and F2(l, 15) = 2.17, ns, for errors. No interaction was found between target condition and distractor condition.
750'
700
Discussion
E
650
600
i t E
550
DETRACTOR SR 500
PI
SR
SU
Target Conditions
25'
B
20•)
O
154
10-
5-
PI
SR
SU
Target Conditions
Figure 11. Mean correct response times (A) and mean error rates (B) for Experiment 3 as a function of the three target conditions (physical identity [PI], semantically related [SR], and semantically unrelated [SU]) and the two distractor conditions (SR and SU). The vertical bars represent standard errors.
p < .012, and errors, Fx(2, 28) = 5.4, p < .01, and F2(2, 30) = 3.47, p < .044. RTs were shorter and accuracy higher when the reference object and the target were physically identical (585 ms and 1.2%) than when the pictures were physically different and semantically related (613 ms and 2.8%), Fa(l, 14) = 6.58, p < .021, and F2(l, 14) = 6.06, p < .026, for RTs. There was no significant difference in the error rate. RTs and errors increased from semantically related pictures to semantically unrelated items (624 ms and 3.1%), F^l, 14) = 5.87, p < .028, and F2(l, 14) = 5.14, p < .037, for RTs, and Fj and F2 < 1 for errors. RTs were longer and the error rate higher when the distractor was semantically related to the reference object than when it was semantically unrelated (615 ms and 2.9% vs. 600 ms and 1.8%), Ft(l, 14) = 4.5, p < .05, and F2(l, 15) = 6.24, p < .023, for RTs, and F^l, 14) = 3.3, p
